Strtok behaviour is defined in the standard(http://pubs.opengroup.org/onlinepubs/009695399/functions/strtok.html) as follows:
A sequence of calls to strtok() breaks the string pointed to by s1
into a sequence of tokens, each of which is delimited by a byte from
the string pointed to by s2. The first call in the sequence has s1 as
its first argument, and is followed by calls with a null pointer as
their first argument. The separator string pointed to by s2 may be
different from call to call.
The first call in the sequence searches the string pointed to by s1
for the first byte that is not contained in the current separator
string pointed to by s2. If no such byte is found, then there are no
tokens in the string pointed to by s1 and strtok() shall return a null
pointer. If such a byte is found, it is the start of the first token.
The strtok() function then searches from there for a byte that is
contained in the current separator string. If no such byte is found,
the current token extends to the end of the string pointed to by s1,
and subsequent searches for a token shall return a null pointer. If
such a byte is found, it is overwritten by a null byte, which
terminates the current token. The strtok() function saves a pointer to
the following byte, from which the next search for a token shall
start.
Each subsequent call, with a null pointer as the value of the first
argument, starts searching from the saved pointer and behaves as
described above.
This means that you could just call strtok two times, and then determine the location just past the \0 of the second substring to get the third part you want.
However, this doesn't seem to be a reasonable way of doing this. It is inflexible, both in dealing with an error (like when the third substring is empty), and with potential future expansions. Furthermore, because of the design of the strtok interface, using it is not thread-safe at all.
It is probably a better idea to hand-code a small lexer/parser that does what you want, or use a tool specifically designed for building lexers (and parsers if needed). I personally have had good experiences with flex for this purpose, but there are other options.