I hear differences throughout the entire ~2:30 length of the song. The most noticeable differences are in sounds with sharp transients.
After finishing the song I exported it without realtime export enabled (because exporting with this option sometimes fails given the large number of plugins used in the project). But when I listened to the MP3 file, it sounded different. Initially I thought the difference was due to poor mixing. So I went back into Cubase, make some more tweaks, re-exported, and listened again. After a few trials of this I realized that it was not me – it was Cubase!
So to see what was happening between the two options, I created some plots that comapre the two exports.
Note, they were made one right after the other – i.e., nothing changed between the two exports. But just to make sure, I repeated the export process three times (as well as the order in which I exported the MP3 files with/without the realtime export option).
The first figure plots MP3-1 (non-realtime export) against MP3-2 (realtime export) for both the left (blue) and right (red) channels. If the two files were identical, you’d see a 45-deg diagonal line. And this is, in fact, what you see when you plot MP3-1 versus itself (that’s the black diagonal line). However, when you plot MP3-1 vs. MP3-2, you get a cloud of points that shows they’re quite different.
To visualize the difference in the waveforms themselves, compare the two plots below (which show the first 250-750 ms of each waveform on the same X/Y scales):
And here’s a screenshot of the project: