Levin, it is a complicated issue for which there isn't an easy answer. Part of the issue is that "block size" doesn't actually measure resource usage very reliably. It is possible to support a much higher volume of typical usage transactions than transactions specifically constructed to cause DoS issues. But if "block size" is the knob being tweaked, then you must design for the DoS worst case, not the average/expected use case.
Additionally, there is an issue of time horizons and what presumed improvements are made to the client. Bitcoin Core today can barely handle 1MB blocks, but that's an engineering limitation. So are we assuming fixes that aren't actually deployed yet? Should we raise the block size before that work is tested and its performance characteristics validated?
It's a complicated issue without easy answers, and that's why you're not seeing straightforward statements of "2MB", "8MB", or "20MB" from most of the developers.