> hmm, yes i query monitoring tables ... how often ? hmm i don't really know (but i thing often). As soon as one query take more than 2 secondes i query the monitoring table to know the number of read to know if the problem is in query optimization (lack of index or wrong plan) or in server overloaded

It really was important info. I've modified your procedures and added simple "select 1 from mon$database" into all of them to create load on monitoring subsystem and reproduced "invalid lock id" bug.

Investigating reasons and hope fix will follow soon.

Thanks,
Vlad