Another problem i found during my tests is that, if garbage collection is deferred to backup/sweep, by executing backup/sweep instead of "select count(*) from table", Firebird will become unresponsive for minutes.

Later today or tomorrow i will file a bug in the firebird tracker, about the problems I have encountered with index garbage collection on varchar index, causing the server to become unresponsive. Because there is a lot of scenarios where we use index on varchar columns that cannot be made sequential and we need to delete old data on a regular basis. In our domain this is social security number, login tokens, oauth access/refresh tokens, guids from third party applications and so on.  

//Thomas