It seems odd to me that the delay is consistently 1.8 sec, even for writes.
I would scan the drive from start to finish using MHDD. You may need to temporarily reconfigure your BIOS for legacy or IDE compatibility mode. MHDD will identify any “slow” blocks, ie those that require more than 500ms to be read. If the resultant bad block map has a checkerboard pattern, then this would tend to indicate a “weak” head.
Alternatively, you could try using dd to copy blocks of 10MB, say, to dev/null, from various zones on the disk. Then look for a pattern in your blktrace log. If there is no pattern, and the slowdowns are truly random, then this will at least eliminate a bad head as a possible cause.
