Jim Allchin looks worried these days. He has good reason to be. Allchin spent an hour last week doggedly trying to rein in the unbridled expectations for Windows NT 5.0 in front of 200 reporters, reviewers and analysts.

By now you already know the drill: NT 5.0 is miserably late. Major features are still under construction. Most of all, even when it's in shape to ship, it won't be the spectacular product whose praises Allchin was singing last September, when the first NT 5.0 beta arrived. It won't be insanely great or perfect - just, well, pretty good.

Allchin should be worried about how far NT 5.0 will fall short of its hype and how late it's running. He's Microsoft's big dog on what's supposed to be one of the company's biggest products. If customers' faith in NT is shaken - well, that's bad news in Redmond.

But Allchin also should be worried about something that goes far beyond spin control. That's the possibility - the likelihood, actually - that NT 5.0 has become a monster.

Every day at 5pm, the NT 5.0 team recompiles the entire operating system. Testers spend the night running it through endless automated tests - stress tests, compatibility tests, tests that simply put NT 5.0 through its paces. In the morning, team leaders decide which bugs will be fixed that day, and the legions of programmers get their marching orders. By 5pm, the changes have been checked in, and the cycle starts all over again.

That's a fine, sensible way to squeeze out the last few bugs at the end of a large project. But this daily build/test/fix cycle has been going on for a year. It's expected to keep going for at least another year.

You recognise the symptoms, don't you? Windows NT 5.0 is hopelessly out of control. It's a classic monster project run amok. It will never really be done - just declared finished someday when Bill Gates gets fed up with waiting for this cash sink to turn into cash flow and sets it loose on the world.

And if you've seen a monster project up close, you already know the only sane response: end it now. Kill it.

Then dredge through the remains for anything you can salvage, and start fresh with a more workable approach - one that chops the project down into manageable pieces, gets them right one at a time and then makes them work together.

The irony is that Microsoft has been preaching the virtues of component development for years. Simplify software development by using our tools to build pieces that can be tested separately and then plugged together - that's the story Microsoft's marketers love to tell developers.

Even NT started with a clean, simple, modular design. That was back in 1991, when Dave Cutler unveiled Microsoft's New Technology kernel ("NT" for short) for a different crowd of reporters, reviewers and analysts.

But apparently that's no longer an option for NT's developers. Instead, they'll watch the days slip by as they pile new features ever higher, hoping that one day they'll have killed enough bugs that, despite the impossible complexity and uncertain quality, Jim Allchin or whoever succeeds him will finally sign off on this monster.

And when that day arrives - after all that time and all those millions of debugging dollars - millions of users will be faced with a Windows NT 5.0 that's late, shaky and, at best, just ... pretty good.

And that's something for Jim Allchin - and the rest of us - to really worry about.