It cracks under this very specific case when one station does massive updating (recalculates few thousand rows in 2-3 tables) and there are lock conflicts in the mean time. There are cases when I can coordinate the sync manually (it`s not needed often, happens only when the internet is down for more than a day or two at a site) - if I let it happen 'in peace' (does the same updating but there are almost no conflicts with other stations trying to write to the database), it goes OK. In fact that`s mostly where I feel the extra RAM - used to be totally impossible to let it happen during work hours, because it held records for too long.

Also, every Sunday I run a SP on the database which sometimes grows the database by up to 1GB  (not related to firebird, there are some errors in the tables with ~30M records which I need to 'heal') - always goes ok but there are no other active connections at the time. I know it`s somewhat of a strange db design and I`ve tried to push optimizations (many of them unsuccessfully). I`ve came to terms with it taking time. What bothers me is it raising a bugcheck.