Update: John Siracusa weighs in
From my perspective, that looks like a tall order, but so far it's looking like a slow news week, and Cristopher Ryan, writing in the Apple Blog, has asked What Could Make OS X 10.7 Great? so I'll have a go.
Without really understanding the issues involved, let me echo Ryan's call for a new file system. Everyone who followed such developments seemed encouraged by the hope that ZFS was just around the corner, when it appeared to be so, and then Oracle bought Sun Microsystems and suddenly we were back in bed with HFS+, with no realistic alternatives on the horizon. Meanwhile, whatever Apple is using for a file system in iOS devices is fairly free to evolve, since iOS apps are sandboxed and only have access to the narrow slice of that system that relates directly to them. Also, being flash memory devices with no hard drives, they have different abilities and requirements as compared with a desktop machine. One possibility for the future is that iOS's refined handling of flash memory might be grafted into HFS+ at about the same time that Macs other than the MacBook Air get a bank of flash memory or a solid state drive to augment their massive (but slower) hard drives. Granted, this probably won't be enough to satisfy those who really understand file system issues.
Another iOS capability that might be brought to Mac OS X is the full-blown touch interface, not just the trackpad gestures supported by the Magic Trackpad (which I love!). (If you think the possibility of touchscreen Macs isn't even on Apple's radar, check out this and this.)
Something not yet implemented in iOS, that I hope to see sooner rather than later is low-level support for machine vision, stereo machine vision to be precise, meaning dual video cameras atop MacBook screens and Cinema Displays, or, perhaps better, a dual camera accessory, also sporting stereo microphones, with motorized pan, tilt, and zoom, as well as automatic focus and aperture, benefitting from the same attention to detail as the original iSight. It would be enough if it could do good quality stereo video recording, editable in iMovie, to begin with, leaving any machine vision applications for the following year.
Mostly, I'd like Apple to continue with the transformative process that was the main selling point of 10.6 over 10.5, making the next version even more coherent, robust, and svelte.