If you are using MPC-HC, you can possibly get a bit lower CPU utilization by using "D3D Fullscreen" and disabling "Accurate Vsync".
If you're trying to play back VP9/WebM videos, you'll get a massive speedup by using MPC-HC v1.7.8 or by manually installing the newest development trunk version of LAVFilters.
In LAVfilters, decoding with Intel QuickSync should be faster than copy-back except with some certain setups in the cutting-edge development versions of LAVFilters or in situations involving fancy subtitles.
in SVP, especially with fast movement at large resolutions, "Motion vectors precision" makes a big performance difference with minimal difference in motion (it makes a larger difference with slow movement at smaller resolutions).
The only other thing is that in SVP, using "Uniform" seems to be a sort-of cheaty way to get better smoothness since CPU usage is barely higher and in my experience the artifacting isn't noticably worse if at all. However, according to this post, you will get more artifacting with "Uniform" if your display refreshrate is not an exact multiple of the source video's framerate (likef 24fps->60hz rather than an exact 3x of 24fps->72hz)
Speaking of display refreshrates, setting your display refreshrate to an exact multiple of the video framerate will definitely improve smoothness and shouldn't result in any increased CPU usage (case in point, 60fps@60hz without interpolation is smoother than 60fps->90hz, but of course 60fps->120hz is smoother than 60fps@60hz). I recommend using MPC-HC's built-in setting for fullscreen to automatically change your resolution and refreshrate based on the video's source framerate. Note that you may need to make some custom resolutions and refreshrates, particularly for something like 72hz - you can make such custom resolutions and refresh rates via the graphics control panel on Nvidia and moderrn Intel GPUs, while AMD Radeon users can use CRU - Custom Resolution Utility.
Personally, when I don't have the CPU performance to use default settings on everything, I first try reducing "Motion vectors precision". If I still don't have enough CPU performance, then I start reducing "Motion vectors grid" (this makes a big difference CPU-wise, but it's also very important for smoothness, but the higher the resolution the less important it is).