Firstly: There’s no Moore’s or May’s Laws (as in scientifically proven laws). I’m surprised people still hang on to these self-fulfilling “laws”, and they should simply be dropped.
Secondly: May’s law doesn’t make any sense at all and is even more nonsense than Moore’s Law. If you run the same software on a faster CPU the software will run that much faster. You very seldomly need to make the software more complex to handle a faster and more capable CPU. Rather, the reason for software becoming slower is because of utilization of the possibilities faster and more capable CPUs provide, which is a completely different story. E.g. graphical vs textual UIs, or video decoding in software vs in hardware etc.
Thirdly: Web-based platforms lend themselves perfectly for massively multi-processor solutions, as the need to store state for such solutions is limited. The key issue is to make also database engines massively multi-processor, but it’s definitely possible.
What’s needed here is a new Web server platform that can allot sessions to many different CPUs, and also a means for allotting database queries to multiple processors and handling of mass storage that is accessed by many processors. Ideally without having to know about this from a software development point-of-view.
Programming-wise there should hardly be any differences anyway, as developers are already used to the state-less paradigm of Web services.
All in all, this is not impossible in any way, and past investments in software development can be retained, even though there might be benefits in developing software specifically for massively multi-processor hardware.