You are probably right with the rounding @vanman , as the String representation of the value for start and end points only goes to 3DP, where as Curve.Length goes further into the decimal.
I might look at this unit issue in the morning and see if deconstruction of the line coordinates further changes anything. 
@vanman moving your added coded inside the Python node by adding the Length value ( as per the observation by @danail.momchilov ) to the String Curve representations should allow for better accuracy when identifying duplicate Curves.
import sys
import clr
clr.AddReference('ProtoGeometry')
from Autodesk.DesignScript.Geometry import *
## For removing duplicate Curves
def cleanLines(curves):
C0 = []
C1 = []
for c in curves:
sL = str(c.Length)
sO = str(c)
sR = str(Line.Reverse(c))
# Added Length to allow for better accuracy identifying duplicates
if sO+sL not in C1 and sR+sL not in C1:
C0.append(c)
C1.append(sO+sL)
C1.append(sR+sL)
return C0
OUT = cleanLines(IN[0])
to awesome. Thank you
Also duplicates can be cleaned with Bimorph nodes.
Thanks Vladimir, I tried Bimorphs node but it takes quite awhile or currently freezes the whole script and I need a shut down. Ewans is working quick as.
Interesting the returns I get. @Ewan_Opie less curves without the length included 
Pretty sure your orginal one is doing the job well and quick though thank you
