The 68,000 number was based on incorrect calculations. By using the files in the directory brown1, and removing the line numbers, I get the following:
[dwin-1210-2:/home/ebender] wc brown1* 5903 62542 397250 brown1_h.txt 5045 58490 329761 brown1_n.txt 5073 58758 329928 brown1_p.txt
The file sizes remain fairly comparable. The following commands count the number of foreign words:
tr ' ' '\012' < brown_h.tag | grep '_FW' | wc tr ' ' '\012' < brown_n.tag | grep '_FW' | wc tr ' ' '\012' < brown_p.tag | grep '_FW' | wc
The result is 33 foreign words in category h, 36 in n and 71 in p. The file sizes are close enough to know just from those numbers that there are many more foreign words in p than in h and n. Normalizing, we find the following:
Category FW/1000 words
---------------------------
H 0.528
N 0.615
P 1.208