custom instruction 'done' signal

I think the done signal from the custom instruction is only used at processor level. Firing the custom instruction will cause the processor to wait until the done signal is there. You can compare it with inserting waitstates for memory access.

If you need a lot of time in the custom instruction, it is better to move the functionality in a custom avalon slave, and poll the slave every now and then on a certain address to look at the state (or use an interrupt). This way, the processor can go one doing other things.