Unless you are running something else besides pgbench on the database, nothing would lock you for two seconds.
Also, the bitmap heap scan is not at fault (it is fast), it is the update itself.
The most likely cause is I/O overload – check the I/O wait time spent by the CPU.
Another (unlikely) possibility, perhaps in combination with the previous one, would be a tiny
shared_buffers, so that the backend cannot find a clean buffer.
CLICK HERE to find out more related problems solutions.