I think trying to keep the number of rules that require context to a minimum is a good idea.  As pointed out in the BIP, using only the timestamp of the block means that the block limit can be determined purely from the block header.

I don't think there is much issue with having a 1MB block following an 8MB block during the activation.

This is inherent in using the timestamps.  It occurs for every block that has a timestamp lower than its parent, but to a lesser degree.

When fees are the main source of income, it does create a slight incentive to use higher timestamps, but that is probably not massive, since it is 2 hours out of the 2 year doubling time.