SGT you are a bona fide genius. I should just listen to you the first time, every time. Thank you for saving my day!Remember what I was saying about not monitoring the current during the time the MOSFET is turned off? The current getting recycled through the diode isn't being measured? (see reply #5).
Well, that's what you are running into.
Have a look at the attached. I yanked out the discrete PNP/NPN gate driver and threw in an LTC4440 high-side driver. It's not the solution for you, because the '4440 has an 80v limit - but it's good just to show you how one could do this (implement a high-side driver so that all of the current through the load is measured, whether sourced from the MOSFET or recycled via the diode).
I removed most of the port type flags for power rails, as they drew attention away from the important I/O signals. I left the +96 port type, because it's HV (even though I lowered it to 75v for the demo).
Even though this configuration works to limit the current, I'm still not terribly fond of it. The MOSFET array keeps getting turned on for short "blips" when the PWM signal goes high; that will result in power dissipation in the module.
The diode you chose is woefully inadequate for the simulation - I left it as it is, but you should change it to an MBRB2545CT. That one isn't rated for the voltage either (~75 volts present at times) or current, but at least it'll have much less of a voltage drop across it (the excessive voltage drop during current limiting results in the boost cap being charged to ~20v).
The components I used for the most part are just dummies to get a simulation going. I will spec better parts before I build anything (which I will probably start asking about in short order). The diode I have in mind is this one:
http://www.mouser.com/ProductDetail/IXYS/MEO550-02DA/?qs=t7yjd2JO/gSLM9NDot8IsA==
200V 582A ~.9V drop