Your shell command line might sometimes work but the problem is precisely that it is not guaranteed to do so. The reason is that the >file part clobbers the very same file that is supposed to be read by the sed -e '1d' file part.

If there was only one process involved, the outcome would be quite predictable. However, since you constructed a pipeline of two processes there is a chance that the first one wins the race and catches a portion of the file before the file is destroyed by the second one. However, as you already observed, you can not rely on that.

To solve that problem you can use a temporary file (like perl -i does behind the scene) or read and write to the file through a single file handle in a single process, which may prove somewhat more difficult to get right.

If you are interested anyway you may want to look up file access modes in perlopentut, specifically +<. You also might find the truncate function useful. The Perl Cookbook has excellent explanations of the different techniques.