The LVM express VI was intended for small data sets, since it is, by default, an ASCII format. As such, it converts the entire data set to ASCII in memory. It is relatively efficient, but still makes a few copies of the data. The ASCII format can easily take up much more RAM than the original binary values.
Due to memory optimization issues, LabVIEW tends to hold onto memory once it is allocated. However, you can probably clear this by writing a one point array.
This also brings up a workaround to your issue. Write the data in smaller increments. For example, if you have 1 million points, write in 10,000 point or 1,000 point blocks. You can set up the express VI to concatenate to the current file. There is a known bug with this method - the number of points in the header will not be updated.
I would urge you to use a binary method, instead, however. You can always use flat binary. National Instruments recommends TDMS for new applications (this is an option on the LVM Express VI), although this option is only available for LabVIEW 8.2 and later (the earlier TDM format has performance issues similar to LVM). For earlier LabVIEW versions, use NI-HWS, available from the driver CD. If you would like more information on the binary file formats, let us know.