Had another play with the overrun fuel cut-off. Schoolboy error it would appear - the MagicJet injectors were bottoming out spectacularly. Ok on idle (just) but as soon as you hit overrun, and/or got a microinjection, it was stuffing in huge dollops of gas, missing, then recovering. (this shows up as lean on an oxygen sensor by the way, when it misfires)
Nice of the ECU to flag absolutely no error for this.

Even nicer of the ECU to set a truly insane minimum opening time (3.8 ms) too.
Special AEB CK8 file for MagicJet injectors
Thanks to Simon P (supercharged Lexus V8 in the Range-Rover) I received a copy of the "special no warranties expressed or impled" CK8 file from AEB. Link here:
http://www.cosic.org.uk/MarkoStuff/lpg/ ... t-rev1.ck8
They've modified it to allow the MagicJet injectors to fire down to 2 ms, not 3.8 ms. Whilst they are officially supported, and bundled in the frontend kits by the integrators, I don't think you stand much of a chance using them with an AEB ECU unless you've got the non-standard config file, thanks to the insane minimum opening time.
(I can understand the logic of not trying to idle below 3.8 ms, if they've been found cheap/nasty/not stable below this by AEB, which is what the emails/disclaimers that Simon P received seemed to state, but you do need the ability not to shovel loads of gas in on the overrun, and the standard config file won't let you do this)
Anyhow - here's the new map and whatnot for an 820 on MEMS:
A few "funnies" that might be useful to know:
ECU options greyed out
Various options in the King software (start hot engine on gas, extra injection cutting) are greyed out when you are connected to the ECU. To select these, you need to load the config from the ECU, save the config file on disk, disconnect from the ECU, choose those settings, save the config file on disk, then reconnect to the ECU, then load the config file to the ECU. Explanations as to how this possibly makes sens welcome on a postcard, but it could save some head scratching!
Microinjections, overrun, and fuel cut
You'll see that my map runs from 1.0 ms to 8 ms. MEMS on the naturally aspirated 2.0 engines injects between 1.0 ms and 8 ms only. Idle is about 2 ms. Mild overrun, the sort that you get lots of transients through in a car park or traffic, is about 1.5 ms. MEMS does micro-injections of 1.0 ms AND in overrun will drop down to 1.0 ms before going to complete fuel cut. I've hard-coded a big fat zero in for 1.0 ms. This kills the microinjections in a way that you have control over, rather than relying on guesswork from the AEB ECU. It doesn't miss them on gas during accel etc.
Being a pikey, I took out 1.25 ms as well, and made 1.5 ms partially fuelled. Making the fuel cut more aggressive basically, as you spend a surprising amount of town driving in that area. (factory fuel cut is over 2600 rpm only, and when completely off the throttle only) Sir who isn't great with a manual gearbox test driving a brand new Rover might not have liked it back in the day, but the rest of us can drive around the transient quite happily. Push it any further and it becomes annoying.
There's an exception at tickover. Too aggressive with the fuel cut here and if you dip the clutch at the wrong moment in a car park you can find yourself in the zero part of the fuel map. You don't get very far with this.
Starting/cranking and idle
Below a certain rpm (I think 400 or so), the MEMS ECU goes into cranking/starting mode. Lots of micro injections and general strangeness that the gas ECU won't appreciate. Don't try to start the hot engine on gas. I've also set it to drop to petrol at 600 rpm (idle is 850) because *if* you're about to stall it, and you happen to dip into the cranking rpm, then it will stall on gas as all that general strangeness kicks in. Pop to petrol and it'll recover fine then switch back to gas again.
General tuning
I found it easier to look at lambda rather than petrol injection times when mapping. Fiddle until you get lambda doing the rich-lean-rich-lean dance the quickest. It'll do this when the petrol injection times are the same on gas and petrol. Much faster than switching between fuels and waiting for the 25 seconds the King ECU demands to elapse.
Idle, overrun at 3000 rpm, 6 ms at 2000 rpm, and 3 ms at 2000 rpm are the most common bins that the dot sits in whilst driving. It's a torquey engine that seems to like being flogged around the 1500-2500 rpm range during accel, then cruises in the 3 ms bin. I kept it stoich up until 7 ms and 4,000 rpm, or 3 ms over 4,000 rpm. Above those loads I leaned it until I'd see the odd "lean" show up on the lambda then added 10% (so that it's always showing rich) and it seems to like this.
At idle with a good oxygen sensor, MEMS switches about 3 times per second. Time to replace if it's slowed down to 1 change per second. This is much quicker than the run-of-mill narrowband oxygen sensor equipped engines of the late 80s/90s, and if you do the blowtorch test on one of the MEMS sensors you'll see that it responds much faster than a generic narrowband. If you've got a car that's misbehaving, check that it hasn't had a universal fitted - they don't work with this engine management system...
http://www.gendan.co.uk/product_O24WIREB.html
Nozzles are 2.8 mm; reducer pressure 1.27 bar. (1.27 +0.05/-0.2 in practice; not the most stable)
Will report back with economy on gas