You make good sense when you support the JTAG features. However, when you have three days in which to complete a project, you really can't build a PCB for those fine-pitch parts. .
Then buy it, or even better, have some on hand.
I have stack of a variety of these: http://www.silabs.com/tgwWebApp...ttools.htm and I just plug "the additional stuff" into them when developing customer special one-offs. Often, I get by with nothing but the devboard. I have devboards as well for socketed varieties of the chips for which I have actual ICEs. If anything is guessed to be made in more than 3 ex, I go straight to dedicated proto PCB.

Now, some of the younger guys can solder to pins spaced at half-a-millimeter, but, quite frankly, I have to put on a special pair of glasses when I have to do that, and it's slow going.
That is why technicians were invented. I have one that can replace a 100 pin TQFP in 10 minutes. Of course, he has all the 'magic' tools. I, myself, have a s(l)ight problem with scope probes.

If that doesn't work out, I can always consider another controller, perhaps one that doesn't tie the async comms rate to the internal oscillator, or perhaps one that uses a clock that works out better. It's just another choice.
I know of no '51 deriavtive that does not "tie the async comms rate to the internal oscillator" and the choice of clock is UART related.

Now, as an example of "getting by", let me tell you that I actually use a 12MHz oscillator in my largest system to satisfy a USB requirement. To achieve precise (enough) baud rates, I PLL the 12 MHz up to 96 MHz before deriving a T4 clock for the UART.
So, having T2, T3 and T4 and a PLL allow me to use a non-UART xtal and still stay well within 0.5% at realistic baudrates (up to 460k).

Erik