1) In modern usage the PECL terminators would be tiny resistor networks.
1a) If the PECL signals are fairly static slow data these may not have to be terminated on a short on-board run. But that leads to the question, why did you convert them to PECL in the first place if they are slow unclocked data?
1b) If the PECL signals are clocks or latch enables etc. you probably want to terminate them.
1c) By Thevenin's Theorem most usual way to terminate PECL will end up being like figure 4 in
http://www.micrel.com/_PDF/HBW/App-Notes/termination.pdf2) One terminator at far end. But it is unusual (see my note below 4) to fan out PECL. If necessary to split a signal multiple ways there are fanout buffers e.g. SY100EP11U, but this would be an unusual use of PECL.
3) Wrong. But note by Thevenin's theorem, a 50 ohm resistor to Vcc-2V, can be replaced by two resistors one to ground and one to Vcc. See 1c above.
4)
http://www.onsemi.com/pub_link/Collateral/MC100EPT622-D.PDFOverall note: Today you are most likely to see PECL or LVDS used on board-to-board or assembly-to-assembly interconnect cables. It'll be translated to whatever's used on the board right at the terminator.