F-Zero X's music was indeed designed for the utmost efficiency, so that as much CPU and RSP power could be used for graphics. No actual synthesis is done, the music is simply decoded and played. Thus we see the size/speed tradeoff in action. Tetrisphere, New Tetris, and Magical Tetris Challenge, on the other hand, just have huge samples and synthesize at a high sample rate, which they can afford since there isn't much in the way of graphics or heavy processing in Tetris games. Ogre Battle is sort of the same phenomenon, it has fairly simple graphics and thus can afford to spend a lot of CPU on sound. It's a matter of balancing the limited power of the N64, not necessarily how efficient the programmers were (there's only so efficient resampling and mixing can get...)
For Phony I have a CPU Usage of 18 % For Ogre Battle 64 - Title 20-22% For Blast Corps Songs 15 % For Golden Eye 64 ONLY 4-7 % For Perfect Dark 7-12 % For F-Zero X 3-6 % For Banjo & Kazooie - Title 13-17 % For Beetle Adventure Racing Songs only 4-7 % For Mario Kart 64 7-9 % For Pilot WIngs 64 3-6 %
Only for Phony and other Songs of Tetrisphere and for especially Ogre Battle 64 are "high" CPU Usage. F-Zero uses very low CPU Ressources...
It's true that the N64 doesn't have a dedicated synthesis chip, the sound hardware is a simple DMA and DAC unit built into the RCP called the AI (Audio Interface). Thus all it can do in hardware is play back PCM data from RAM. Most games (dare I say all?) generate the sound via microcode on the RSP, which is designed for processing large amounts of data. The RSP is also used for geometry transformations, games load microcode onto it that takes the higher-level display list generated by the game and processess it into the low level "this triangle at these pixels with this texture" type of commands that the RDP can use directly. The design of the N64 is such, however, that you can generate audio and graphics on the CPU or RSP in any manner you choose. I seem to recall reading about Indiana Jones and the Infernal Machine that the developers designed the sound system so it would do audio generation on the CPU if the RSP was too busy with graphics. My NES emulator, Neon64, never touched the RSP for sound, and initially it actually used the RSP for rendering instead of the RDP (and very early on everything was done right on the CPU).