UPDATE 1-Kenyan referendum "Yes" vote up to 66 pct - poll
* "Yes" majority in in all eight provinces
* Kenyan financial markets expecting it to pass
* 44.5 pct say now have doubts about their church leadership
NAIROBI, July 30 (Reuters) - Two-thirds of Kenyans intend to vote in favour of a new constitution in a referendum on Aug. 4 and 87 percent of those surveyed will accept the result whichever side wins, an opinion poll showed on Friday.
The poll by Strategic Research shows the "Yes" vote has risen from 62 percent since the pollster's previous survey published on July 16 as more people make up their minds, while the "No" camp has remained stuck at 20 percent.
While Kenyan financial markets have largely discounted a "Yes" win, analysts say at least two-thirds need to vote in favour to give the new constitution credibility and there remain concerns about violence, especially in the Rift Valley.
The survey showed the "Yes" camp would have a majority in all of Kenya's eight provinces.
The contest would be tightest in the Rift Valley -- the epicentre of Kenya's 2008 post-election violence -- where 51 percent said they were in favour of the new constitution and 34 percent against.
The leaders of the coalition government have been backing the new constitution while cabinet minister William Ruto and former President Daniel arap Moi have been spearheading the "No" campaign.
Christian church leaders are also against the constitution due to a clause allowing abortions on medical grounds and recognition of Islamic courts dealing with divorce and inheritance.
The poll showed, however, that 68 percent of Catholics surveyed would vote in favour, 63 percent of Protestants would say "Yes" while 74 percent of Muslims backed the draft.
Asked how the position of church leaders has affected their relationship with their church, 44.5 percent said they had developed doubts about their church leadership and 35 percent said they liked their church even more.
The random nationwide survey of 2,400 people was conducted on July 26-28. The sampling error was 1.6 percentage points at a 95 percent confidence level. (Reporting by Richard Lough; Editing by David Clarke)