No idea why you're getting down voted but that was my thought as well.

With immediate mode you don't have to construct any widgets or objects. You just render them via code every frame which gives you more freedom in how you tackle each UI element. You're not forced into one widget system across the entire application. For example, if you detect your tooltip code is slow you could memcpy all the strings in a block of memory and then have tooltips use an index to that memory, or have them load on demand from disk, or the cloud or space or whatever. The point being you can optimise the UI piecemeal.

Immediate mode has its own challenges but I do find it interesting to at least see how the different approaches would tackle the problem