Please explain this a bit more...

> Why should the kernel or the userspace care whether its allocation is contiguous or not?

In one word - performance.

Also

> Besides having larger contiguous areas that could be used i.e. for DMA transfers or some IPC..

I believe with certain bits of hardware you need to substitute the word "could" with "must".

At the end of the day, spending ten otherwise idle cycles to gain 100 cycles when under pressure is money well spent, and I get the impression the gains are much bigger than that.

Quite possibly not B.Sc. level though.

Cheers,
Wol