Mmm, another Rachmaninoff disputed accidental. I have the Boosey & Hawkes edition, in which the note in question is printed as a C#. (Well in fact it isn't printed as anything but there's a C# earlier in the bar). As written it's an odd sound but not an entirely horrible one, but I think it's probably an early mistake that's become permanent. Certainly it would be very easy to miss. Anyway, the second half of b.32 parallels the second half of b.11 where there certainly isn't a corresponding G#.
On the recording I have (Richter) it's not completely clear, which suggests to me that he's playing the natural - if he played the sharp, you'd notice.