Thanks for getting back to me.
Yeah, I fixed the discrepancy in size of basic sketches (also made LTO enabled by default), that's in the latest board manager release. If the difference was significant, though, I think it was the LTO.
Next release will include a new tinyNeoPixel builtin library, which is integrated with the core to minimize flash use (comes in two versions, tinyNeoPixel, which implements the full Adafruit_NeoPixel functionality, and tinyNeoPixel_Static, for which you have to statically declare the array to hold the pixels, and you lose the ability to change the length at runtime and have to manually call pinMode(pin,OUTPUT), but in so doing saves yet more flash; if you can replace all pinMode calls with DDRx register writes, the _Static version of the library will not pull pinMode back in for the single call to enable, saving approx 100-some-odd bytes).