I’d stick with that, Ambershee is right in terms of the issues with TAA. For video rendering do what you’re doing now.
The reason for MSAA not always working so well in a Deferred Renderer isn’t so much because of the technique, it’s because the GBuffer that comes out from a Deferred Renderer is usually SO badly aliased that it’s beyond the help of multisampling, unless you use a high multiplier. The Quality vs. performance at that point just makes the Deferred solution almost useless.
Also, I can’t remember if the aliasing works before or after the GBuffer has been rendered but, there is only a certain size you can go to when rendering the GBuffer (I think the limit for Unreal is 16K). You can render it at a higher resolution, but there is a cap in DX / OGL that you’ll hit eventually anyway, but the main issue is that each step up costs you 4X the rendering time (ish), so MSAA has an upper limit anyway on what multiplier you can use, which probably still isn’t good enough. At least, that’s how I think it works.
I think it was Joe Graf that did a presentation on why they settled with TAA and didn’t use MSAA… I’ll try and dig it out.