Under what circumstances does this happen?

Assuming the request is on a TCP connection (which is the usual  
mechanism), you'll get a zero byte read any time the connection has been  
closed.  In the normal case, this happens when the data's all been sent  
and you've read everything there is to read.  But I suppose you could have  
an HTTP server that's closing the connection prematurely or something like  
that.  The ContentLength is just what the HTTP server said it would send;  
there's no guarantee when you get that value that there is actually that  
much data ready and waiting for you to receive.

It may be difficult to diagnose the problem if you can't come up with a  
reproducible scenario.  If you can, then the first thing I would do is run  
a network analysis program (like Wireshark, for example) and see what's  
really going on with the connection.  If the connection is in fact being  
closed before you've read all the data (which seems to me to be the most  
likely explanation), then you may be dealing with a server issue (or  
possibly even an ISP issue...after the whole Comcast/BitTorrent fiasco, I  
have little faith that all ISPs are leaving the data stream alone all the  
time).

Before you do all that, you may of course want to just double-check your  
code and make sure that your logic is simple and doesn't have any way that  
it might accidently read something without including it in the final data  
count.  For example, some sort of odd threading implementation where data  
can be lost due to synchronization problems between threads.  I think a  
person would have to go to some effort to write code like that, but I've  
seen worse.  It could happen, and you should double-check and make sure  
that's not going on here.

Pete

It sounds weird.

The docs are rather clear:

#Return Value
#The total number of bytes read into the buffer. This can be less than
#the number of bytes requested if that many bytes are not currently
#available, or zero (0) if the end of the stream has been reached.

Is the file in question public available so we can try ?

Arne

I don't know why it is dropping ,but if the web-server supports it,
you could attempt to resume by adding a RANGE header to the request
and re-sending? (i.e. from byte {large number} onwards...)

Marc

Would the amount that you actually can download correspond to about four
megabytes?