Are you trying to parse real XML, or just to build a tool that can handle some small subset of it? The reason I ask is that actual XML can contain a lot of oddities that your code above will not handle property. For instance, XML allows attribute-value pairs to have spaces around the equal sign, like this:
<Root id = "456990">

It looks like your code would choke on that, though. Unless you are absolutely sure that your input data, now and forever, will not contain anything but the ultra-strict subset of XML that your code will support, I would urge you to use XML::Parser. (I believe its internals are written in C, so it's actually quite fast; have you benchmarked it on your specific documents to see if it will meet your needs?)

I know you mentioned your criterion for efficiency is execution speed, and that you don't want to use a separate parser, so maybe I should just butt out. It's just that years of working with HTML and more recently XML have taught me to be extremely cautious. Building a parser that really respects the specs is a non-trivial task, and I'd hate to see fragile code go into production and then have to be torn out later for maintenance, when a perfectly good module is already available to do the task you intend.