I think the Ruby-only shtick is that they want to do one thing and do it well. There's enough of a Ruby community that they can serve that niche and do it well.
There's nothing so exceptional about what they're doing that it couldn't be done in Python with what is available (the exceptional part is that they execute well and are highly reliable). Basically, the benefit of Heroku is that they can keep extra hardware on hand for spikes and there's a decent assumption that, say, your site spikes to 300% of normal once a week, but that not all sites will be spiking at the same time. Therefore, you get more efficient utilization of resources.
But you could do this yourself - it would just mean that you would need to buy all the capacity to handle spikes while Heroku works off the assumption that part of that spike capacity can be shared since not all sites will spike at once.
> decent assumption that, say, your site spikes to 300% of normal once a week, but that not all sites will be spiking at the same time.
This reminds me of something.... what is it... oh yeah, the economy. It's probably a better assumption than some of the junk that went on to tank the economy, but I'd still be interested in learning more about spike distributions. Amazon probably has some good data.
There's nothing so exceptional about what they're doing that it couldn't be done in Python with what is available (the exceptional part is that they execute well and are highly reliable). Basically, the benefit of Heroku is that they can keep extra hardware on hand for spikes and there's a decent assumption that, say, your site spikes to 300% of normal once a week, but that not all sites will be spiking at the same time. Therefore, you get more efficient utilization of resources.
But you could do this yourself - it would just mean that you would need to buy all the capacity to handle spikes while Heroku works off the assumption that part of that spike capacity can be shared since not all sites will spike at once.