When WPML was free, it had usage statistics on WordPress.org.
If you combine the total downloads for WPML and several other multilingual plugins on W.org, you'd get something like 50,000 active sites (to calculate this, you should look at download peaks during update days and divide by the percentage of that version).
WordPress 3.2 is currently at around a million downloads and 3.1 got around 15 million.
So, we're actually talking about something like 0.3%.
Most WordPress downloads are for blogs. I think that the total active WordPress installations is around 4 million (that's a number for a different discussion). So, this brings the total number of multilingual WP sites close to 1%.
If multilingual support went to WP core, that number would have risen considerably. However, the overall number of WP users would drop, as the software would become much heavier and more complex - a bit like Drupal is today.
I think that a fair start would be to include proper support for choosing a single language easily.