Eric,
This gets interesting - haven't done the screen grabs yet - I'll do those next. But I did something more in-depth. Important background data - I'm printing with the Epson 3800 using 13*19 inch Ilford Gold Fibre Silk and a custom profile (a paper and inkset combination which puts far less demand on soft-proofing than would a matte paper). I'm working with challenging high contrast images - huge Chinese lanterns lit-up at night by the shore of Lake Ontario. Exposure at capture and then luminosity settings in LR had to be done carefully to get a satisfactory balance between dark tones, mid-tones and highlights, while avoiding clipping important highlight detail.
I decided to conduct two parallel workflows (A) Straight through in LR2 simply activating the print module, setting it all up for the printer, paper size, margins, choice of paper profile, "Standard" Output Sharpening, 300 PPI and push Print. It did a very good job even without soft-proofing. Export the same image to PSCS3 and once there, activate soft proof, make the compensating Curves adjustments, implement PK Inkjet Output Sharpener for 300 PPI and push Print.
Comparing the two images, I couldn't really say that one is overall decisively better than the other in any respect. There are minor differences of tonality in different parts of the image - most likely on account of how I curved for the soft-proof; sharpness was pretty much the same. Needless to say, the CS3 route is more labour-intensive. So bravo to Lightroom.
The advantage of this methodology, while not a fully consistent treatment, is that it replicates my characteristic workflow in each application. To be fully consistent, I printed the CS3 version straight as received from LR2 with no tweaking at all, save for PK Output Sharpening. The two prints - the one printed from LR2 and the other from CS3 un-tweaked after exporting from LR are virtually indistinguishable.
Of course this is a sample of one. I may try several more comparisons over the next while (on smaller less costly sheets of paper!) to see whether these observations hold. But so far it looks as if there is a high degree of final results consistency regardless of this differing display appearance observation, which I shall test with screen grabs as you suggest.
Cheers,
Mark