I can for Oracle, which I'm most familiar with. Its approach adds much of complexity both in implementation and configuration/administation. DB engine must ensure integrity and consistency of data in different DB files in scenarios like backup, restore, crash recovery, DDL etc. DBA have to monitor and tune undo space according to database growing rate and workload. If not done properly, long-running SELECT can end up receiving "Snapshot too old" error, which means DB engine could not reconstruct all row data needed as seen at certain point of time.

Rollbacks (canceling a transaction) on other systems usually take longer because it physically needs to rewrite the "undo log" back to the data pages.