The other problem is holding down small transformers to pass the shake test - it's a pain having to use wire ties in production.

Transformer given in schematic is a very small 1VA type from 'Block', to be soldered directly on PCB. Transformer is hold by a total of 6 solder pins. Should be enough to pass the shake test, what do you think?
Besides providing isolation, transformer shows another advantage: Transformer contributes to heat dissipation with only 0.6W!

As I mentioned in another post, the Sharp pc900v has an input current of just over 1mA.

Have had a look into datasheet. Threshold levels can be 0.3mA and 4mA, when tolerances are maximal. So, I would recommend to use signal current of 5mA for turning on the opto LED. Gives a bit headroom against changes of constant current source.

For the constant current source a n-channel depletion MOSFET would be nice. Something like DN2540. But it's hard to find a depletion MOSFET with maximum drain source voltage of more than 500V. Consequence is, that much filtering is necessary to guarantee the absence of unsane voltage spikes during surges and bursts.

Don't forget to have very good cooling for the transistor used in the constant current source. I once suffered from relevant drift of 'constant' current with heat-up period of according transistor...

Kai