James,
yes, you're right, that's what 'solved' the problem. Tests with about 200,000 records and olb:refresh took up to 3 seconds per dbseek, whereas olb:select(0) and select(1) only take up to 0.2 seconds at most.
But indeed, we are already considering, if it's useful to wait about half a second, if there follows another keystroke, and if not, then do the dbseek.