Sure, frame smoothness is important, but did you read what I said? I said FPS is a valuable tool, but you have to understand what it does. It's like peak horsepower for cars. It's a valuable tool for quickly (and roughly) assessing the performance of a car, but ultimately it's but one point on the RPM band among a host of other variables that impact said performance.
Everything you say about FPS applies to these 99th percentile graphs that you guys seem newly obsessed with. They tell you one small thing, and don't necessarily indicate much about smoothness or framerate. If you have one section where there's a sudden dip, 99th percentile frame latency tells you nothing about the experience as a whole. Again, you want the whole story? You look at a histogram and a time graph of frame latency.
Most people don't have the time/inclination to do that, though, so a combination of a few statistical values works well for giving you a good idea of what's going on. Hey, maybe instead of trying to champion a single value (FPS or 99th/whatever percentile) and ignore all others, we could take a traditional statistical approach and use the mean (FPS), median, variance (this is where your percentile stuff comes in), and bias (distribution bias) to look at the distribution of frame latencies. Nah, that makes too much sense and is too inline with traditional analysis in every field ever.