Check to see what happens if you copy a sample to another sample voice space. If making a copy of the whole "here is my sample" sample doesn't use up any more RAM, then the ASR10 just makes links to the actual sample in RAM. Then, you can set the start and end points of the each of the sample "copies". If you copy it 3 times, then you should be able to truncate each sample voice copy to each single word. Then have a look to see if it used up any significant RAM in doing so. If not, then it is using the single sample data and processing it separately for each sample voice. That's what my Casio FZ-1 does.
Casio: FZ1, CZ101, PG380, VZ10M, SK8, RAP10, DP1 drum kit, DH100, CDP200
Yamaha: TX81z, TX802, SHS10
Roland: SH101, DR660, MS1, PM16, TR33
Others: Korg MicroKontrol & ES-1, Alesis MMT8, Emu ESi2000 & CF card reader, Evolution MK461C, BCR2000