The simple answer is: we don't know yet. We actually had a conference call with some folks from Canonical a few weeks ago discussing this very issue. At present, it seems like there are three options (I'm taking these points mostly from the minutes from that call):
1. Port everything to Qt and use Mir (because GTK+ might not work under pure Mir -- they've been backing Wayland so far). This would require an IMMENSE amount of work and would be tantamount to rewriting almost every line in our codebase.
2. Help port GTK+ (or make an extension/fork of it) to Mir. This might honestly be less work than #1, but elementary would be at a specific disadvantage because most of our developers (speaking for myself as well) don't really have the skills needed for something like this.
3. Do nothing and use XMir until we absolutely must switch.
We don't have an official or internal consensus on this point yet, to my knowledge. However, I would personally tend to think we'll be leaning toward #3, depending on how XMir shapes up.
This is definitely an area of interest and importance for us, and it's something that will undoubtedly continue to be discussed as we begin the L+1 cycle and beyond.