It would be most interesting to hear your thoughts on printing via the traditional Epson driver and the ABW (with modifications as noted above). Mark Segal has done some comparisons in his recent reviews. I assume your custom icc profile (the 3k patch set) contains a B/W step gradient to assist in B/W printing. This is pretty much what I do when I prepare profiles using ArgyllCMS (51 step B/W patch set).
Alan
Actually, I was pleased to see that my normal A3+ profile making target measured on an i1 iSis XL works just fine (~3k) and that making a special one for B&W (normal colour set (~1k) plus 0-255 grey and near greys gave very little improvement if any, with i1profiler - there are not the adjustment options available in i1profiler when making profiles to get any real benefit (life has consistently been too short to get to grips with argyllCMS- YMMV)
Whilst I've lots more prints to try, I'm finding that ABW works best on shiny papers (PK) but with matt papers such as Innova IFA-11 (-very- similar to Epson HPN) I'm leaning towards using my own colour icc profile with a simple PS curve (only 3-5 points specified).
51 points (scanned with i1iO2 3x averaged) gives a good solid data set to see variation of linearity, but a simple curve works just fine - making dozens of prints and actually looking at them is is a powerful reminder that chasing the numbers can be rather futile after a while.
This is an ACV curve for IFA-11 - one of the more complex ones. Curve points are entered manually from a normalised set of readings of a 51 step wedge (I just paste the saved measurement data into a spreadsheet, to create a set of 'L' values normalised to 0-100)
I'm minded to wonder if more complex solutions seem more disposed to those not actually wanting to print photos ;-) :-)
But seriously, an 'excess' of measurements can exacerbate residual measurement errors and minor variations in linearity, producing what might seem more accurate solutions, but when combined with the vagaries of 8 bit curves produce worse looking prints - every time I did a print with a curve/profile/setting, I looked first at the bullseye target on the test print. Several good looking solutions showed up slight banding.
I was asked why I wanted a linearised print output, since you could say that perceptually, there should be curves in the response... From my POV I want to start at linear and add those curves based on image content, paper type and viewing conditions, not have them 'baked in' - it's much like my approach to print sharpening for larger images - I want the option to have absolutely none at all in parts of the image.