Is there any scientific/empirical proof on this? I indent 4 spaces, no tabs, but I would like to see the real rationale there, not just the PEP-8 "rules".

There is this: "Program indentation and comprehensibility (1983) by R J Miara, J A Musselman, J A Navarro, B Shneiderman"

http://www.cs.umd.edu/~ben/papers/Miara1983Program.pdf

I'd classify it as an attempt rather than something decisive but whatever floats your boat.

p866:

  The level of indentation that seems to produce optimal 
  results in comprehension is between 2 and 4 spaces; as the 
  number of spaces increase, the comprehension level 
  decreases.

On the other hand, the sample program hard parts that were indented nine times (!) in places. It's an interesting read, but I'd not try to use it in a discussion.

Personally, I don't care; just pick one (hopefully one that matches what most people are using with that language, eg. 2 spaces for Ruby), and get on with it.

Yup, they seem to have used an overly indent-happy indentation scheme. Two separate indents: he 'begin' after the 'for' and for the block body is superfluous. This raises the concern that their results are biased towards smaller indents.

Zero spaces makes indentation harder to spot than infinity spaces.

Infinity spaces gives less space for code than zero.

Conclusion: use as wide an indentation as possible while still keeping enough space for the code itself.

Is there nothing to be said for consistency?

When I suggest using that algorithm to pick an indent size, I mean to use long-term averages to select a standard, not to indent every line as much as possible individually :P

No. Indent space is just an opinion/preference/habit.

I prefer 2 space indents (tabs)

you're getting used to it within few months.. (currently we have tabs though)